Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
IEICE Technical Committee Submission System
Conference Paper's Information		 Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2021-03-05 13:00
The Relation between Sensitivity and Maximum Lyapunov Exponent when Sensitivity Adjustment Learning is Applied to Layered Recurrent Neural Networks
Takuya Ejima, Yuuki Tokumaru, Kastunari Shibata (Oita Univ.) NC2020-69
Abstract (in Japanese) (See Japanese page) 
(in English) We have proposed a local learning method named "sensitivity adjustment learning (SAL)". The sensitivity, which is adjusted in SAL, is a local index defined in each neuron as the magnitude of the output gradient with respect to the input. In the previous study, we found that SAL can control the global dynamics of non-layered recurrent neural networks (RNNs). In this paper, we observed the relation between the sensitivities and the maximum Lyapunov exponent, which shows the network chaoticity, for several cases of layered RNN when SAL was applied. The results show that the sum of "log sensitivity", which is the natural logarithm of the average sensitivity over all neurons in each layer, corresponds reasonably well to the maximum Lyapunov exponent until the dynamics get chaotic regardless of the number of neurons or connection rate. This suggests that by setting the target of "sensitivity" to be 1.0, SAL can adjust the maximum Lyapunov exponent to 0, which indicates the "Edge of Chaos".
Keyword (in Japanese) (See Japanese page) 
(in English) recurrent neural network / chaos dynamics / sensitivity adjustment learning / sensitivity / edge of chaos / / /  
Reference Info. IEICE Tech. Rep., vol. 120, no. 403, NC2020-69, pp. 151-156, March 2021.
Paper # NC2020-69 
Date of Issue 2021-02-24 (NC) 
ISSN Online edition: ISSN 2432-6380
Copyright
and
reproduction		 All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034)
Download PDF NC2020-69

Conference Information
Committee NC MBE  
Conference Date 2021-03-03 - 2021-03-05 
Place (in Japanese) (See Japanese page) 
Place (in English) Online 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Neuro Computing, Medical Engineering, etc. 
Paper Information
Registration To NC 
Conference Code 2021-03-NC-MBE 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) The Relation between Sensitivity and Maximum Lyapunov Exponent when Sensitivity Adjustment Learning is Applied to Layered Recurrent Neural Networks 
Sub Title (in English)  
Keyword(1) recurrent neural network  
Keyword(2) chaos dynamics  
Keyword(3) sensitivity adjustment learning  
Keyword(4) sensitivity  
Keyword(5) edge of chaos  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Takuya Ejima  
1st Author's Affiliation Oita University (Oita Univ.)
2nd Author's Name Yuuki Tokumaru  
2nd Author's Affiliation Oita University (Oita Univ.)
3rd Author's Name Kastunari Shibata  
3rd Author's Affiliation Oita University (Oita Univ.)
4th Author's Name  
4th Author's Affiliation ()
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2021-03-05 13:00:00 
Presentation Time 25 minutes 
Registration for NC 
Paper # NC2020-69 
Volume (vol) vol.120 
Number (no) no.403 
Page pp.151-156 
#Pages 6 
Date of Issue 2021-02-24 (NC) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan